EXO Labs
Open-source software to run large AI models across clusters of consumer devices.
Updated April 2026
Overview
- Website
- exolabs.net
- Headquarters
- London, UK
- Segment
- Edge & Distributed Inference
Product overview
EXO Labs builds Exo, an open-source platform that connects devices like MacBooks and Mac Minis into AI clusters for distributed inference of frontier LLMs such as Llama 3.1 405B using pipeline and tensor parallelism. It supports OpenAI-compatible APIs and automatic device discovery, enabling local inference runs without cloud dependency. Used by companies for private AI inferences and by researchers for fast experimentation; stands out by optimizing for consumer hardware constraints unlike datacenter GPU setups.
Revenue model
Open-source core is free; venture-funded startup with undisclosed private investment. Revenue expected from enterprise on-premise AI solutions and bespoke installation services; no public pricing tiers or API costs currently disclosed.
Moat
- Proprietary Technology
- Scale Advantages
- Data Flywheel
EXO Labs' key competitive moat is its proprietary EXO Gym platform, which simulates distributed AI training on a single machine, drastically lowering barriers to research and accelerating discovery of efficient algorithms like DiLoCo for large language models.
Headwinds
Consumer hardware limitations may not be sufficient to run truly frontier models effectively, limiting market adoption.